In message <20050825185052.GB17624@netbsd.org>Bill Studenmund writes

>I assume this is over the loopback interface? If so, you might try
>lowering the MTU on the loopback interface. [...]

>If I drop the loopback mtu from 33192 to say 18000, the issue goes away.

Hi Bill,

That's a bug. I dont recall if I committed SIOC[GS]IFMTU for lo(4) to
NetBSD or if someone else did it independently, but I've been using it
since NetBSD-1.3, to measured ttcp throughput over lo(4) as a function
of MTU size.

Does [n]ttcp mibehave the same way as your app?